1.1 AWS MLA-C01 Blueprint, Format & Strategy
Key Takeaways
- The AWS Certified Machine Learning Engineer — Associate (MLA-C01) exam consists of 65 total questions (50 scored, 15 unscored pretest items) administered over a 130-minute testing window.
- Scoring operates on a scaled system from 100 to 1,000 with a minimum passing threshold of 720 points, utilizing compensatory scoring where strengths in one domain can offset weaknesses in another.
- The exam blueprint is partitioned across four operational domains: Data Preparation for ML (28%), ML Model Development (26%), Deployment and Orchestration (22%), and Monitoring, Maintenance, and Security (24%).
- MLA-C01 replaces the retired MLS-C01 Specialty exam, shifting the curriculum away from theoretical mathematical ML derivations toward production MLOps, SageMaker Pipelines, Feature Store, Model Monitor, Clarify, and Bedrock/GenAI engineering.
- Effective exam pacing requires an average allocation of 2.0 minutes per question, utilizing a two-pass strategy to secure quick architectural wins before tackling multi-paragraph operational scenarios.
1.1 AWS MLA-C01 Blueprint, Format & Strategy
The AWS Certified Machine Learning Engineer — Associate (MLA-C01) certification validates end-to-end technical competence in designing, building, deploying, monitoring, and maintaining production-grade machine learning (ML) and artificial intelligence (AI) workloads on Amazon Web Services. Unlike legacy examinations that focused heavily on mathematical ML theory or generic data science modeling in isolated Jupyter notebooks, MLA-C01 assesses real-world MLOps (Machine Learning Operations), data pipeline engineering, feature store design, automated workflow orchestration, model governance, cost optimization, and secure inference architectures.
Achieving this credential certifies that an engineer can operationalize ML models into scalable, secure, highly available, and cost-effective cloud architectures using AWS native services such as Amazon SageMaker, AWS Glue, Amazon Bedrock, AWS Lake Formation, and Amazon EventBridge.
1. Exam Logistics, Scoring Mechanics & Format
The MLA-C01 examination adheres to strict AWS certification standards regarding delivery, question distribution, and scoring algorithms.
+-----------------------------------------------------------------------------+
| AWS MLA-C01 EXAM SPECIFICATIONS |
| |
| [EXAM CODE] ---> MLA-C01 |
| [TARGET AUDIENCE] ---> ML Engineers, MLOps Specialists, Data Engineers, |
| Cloud Developers with 1+ years AWS ML experience |
| [TOTAL QUESTIONS] ---> 65 Questions (50 Scored + 15 Unscored Pretest Items)|
| [TIME DURATION] ---> 130 Minutes (2 Hours 10 Minutes) |
| [PASSING SCORE] ---> 720 / 1000 (Scaled Score: 100 to 1000) |
| [EXAM COST] ---> $150 USD (or 50% discount voucher from prior AWS) |
| [DELIVERY METHOD] ---> Pearson VUE (Testing Center or Online Proctored) |
| [VALIDITY PERIOD] ---> 3 Years from passing date |
| [ESL ACCOMMODATE] ---> +30 Minutes available for non-native English |
+-----------------------------------------------------------------------------+
[!IMPORTANT] MLA-C01 Retirement & MLA-C02 Transition (as of August 2026): AWS is updating this exam. Registration for the updated version (MLA-C02) opens September 1, 2026, and the last day to take MLA-C01 in English is September 28, 2026; the Korean, Japanese, and Simplified Chinese versions of MLA-C01 remain available until MLA-C02 reaches general availability. Before scheduling, confirm which exam version you are registering for and review the MLA-C02 exam guide when published. This guide teaches the current MLA-C01 blueprint (exam guide version 1.0), which remains fully valid for exams taken before the transition dates.
Scaled Scoring & Pretest Questions
- Scaled Scoring Model (100–1,000): AWS converts your raw performance across scored items into a standardized scaled score. The minimum passing threshold is 720.
- Compensatory Scoring: The exam does not require you to achieve 720 in every individual domain. A strong score in Domain 1 (Data Preparation) can compensate for a borderline score in Domain 3 (Deployment & Orchestration), provided your aggregated scaled score reaches or exceeds 720.
- Unscored Pretest Items: Out of the 65 questions presented during the exam, 15 items are unscored trial questions used by AWS psychometricians to validate question fairness and statistical difficulty for future exam revisions. You will not know which questions are scored versus unscored; therefore, you must treat every question with equal diligence.
[!IMPORTANT] No Negative Marking: There is zero penalty for incorrect guesses on the AWS exam. Never leave any question unanswered when your exam timer expires.
2. The Four Exam Domains & Official Task Statements
The MLA-C01 blueprint is divided into four distinct technical domains, reflecting the end-to-end lifecycle of an ML workload.
+-----------------------------------------------------------------------------+
| MLA-C01 DOMAIN WEIGHTING BREAKDOWN |
| |
| +-------------------------------------------------------------+ |
| | DOMAIN 1: Data Preparation for Machine Learning | 28% | |
| +-------------------------------------------------------------+ |
| | DOMAIN 2: ML Model Development | 26% | |
| +-------------------------------------------------------------+ |
| | DOMAIN 3: Deployment and Orchestration | 22% | |
| +-------------------------------------------------------------+ |
| | DOMAIN 4: ML Monitoring, Maintenance & Security | 24% | |
| +-------------------------------------------------------------+ |
| | TOTAL | 100% | |
| +-------------------------------------------------------------+ |
+-----------------------------------------------------------------------------+
Detailed Task Statement Breakdown
Domain 1: Data Preparation for Machine Learning (28%)
- Task Statement 1.1: Ingest and store data: Ingesting batch, micro-batch, and streaming data using Amazon S3, Amazon Kinesis Data Streams, Amazon Kinesis Data Firehose, and Amazon Managed Streaming for Apache Kafka (Amazon MSK). Selecting appropriate S3 storage classes (S3 Standard, S3 Standard-IA, S3 Intelligent-Tiering) and high-throughput training storage (Amazon FSx for Lustre integrated with S3 buckets).
- Task Statement 1.2: Transform data and perform feature engineering: Implementing distributed ETL transformations with AWS Glue (PySpark), AWS Glue DataBrew, and Amazon SageMaker Data Wrangler. Managing centralized, reusable feature definitions using Amazon SageMaker Feature Store (configuring Online Store for sub-10ms inference and Offline Store in S3 with Parquet format for point-in-time training joins without target leakage).
- Task Statement 1.3: Ensure data integrity and prepare data for modeling: Performing statistical data profiling, handling missing values, detecting class imbalance (SMOTE, random over/undersampling, class-weighted loss), managing schema drift with AWS Glue Data Quality / Deequ, and orchestrating human-in-the-loop annotation workflows with Amazon SageMaker Ground Truth.
Domain 2: ML Model Development (26%)
- Task Statement 2.1: Choose a modeling approach: Selecting optimal built-in algorithms in SageMaker (e.g., XGBoost, Linear Learner, DeepAR, BlazingText, Random Cut Forest, Object Detection, IP Insights, K-Means), authoring custom training containers via SageMaker Script Mode (PyTorch, TensorFlow, Scikit-learn), Bring Your Own Container (BYOC), and leveraging Foundation Models via Amazon Bedrock and SageMaker JumpStart.
- Task Statement 2.2: Train and refine models: Configuring SageMaker Training Jobs, reducing compute expenses by up to 90% via Managed Spot Training with S3 checkpointing (
checkpoint_s3_uri), setting up resilient GPU training clusters with SageMaker HyperPod, executing distributed data and model parallelism (SageMaker Distributed Data Parallelism / PyTorch FSDP), and automating hyperparameter optimization (HPO) with Automatic Model Tuning using Bayesian optimization and Hyperband strategies. - Task Statement 2.3: Analyze model performance: Calculating domain-specific evaluation metrics (Precision, Recall, F1-score, ROC-AUC, PR-AUC, RMSE, MAE, R², ROUGE, BLEU), tracking trial parameters and artifact lineage using Amazon SageMaker Experiments, and managing versioning, model cards, and manual/automated promotion gates via Amazon SageMaker Model Registry.
Domain 3: Deployment and Orchestration of ML Workflows (22%)
- Task Statement 3.1: Select deployment infrastructure based on existing architecture and requirements: Choosing between SageMaker Real-Time Endpoints (persistent low-latency), Serverless Inference (intermittent/unpredictable traffic with cold-start tolerance), Asynchronous Inference (large payloads up to 1 GB, long timeouts up to 1 hour, auto-scale to zero), Batch Transform (high-throughput offline batch scoring), Multi-Model Endpoints (MME for cost-efficient hosting of hundreds of homogeneous models), and Multi-Container Endpoints (MCE / Serial Inference Pipelines).
- Task Statement 3.2: Create and script infrastructure based on existing architecture and requirements: Implementing zero-downtime production rollouts using Blue/Green deployments, Canary deployments, Linear traffic shifting, and Shadow deployments (validating performance under live traffic without impacting client responses). Configuring Application Auto Scaling policies (Target Tracking, Step Scaling) and selecting hardware acceleration (AWS Inferentia / Trainium vs. NVIDIA GPUs vs. CPU instances).
- Task Statement 3.3: Use automated orchestration tools to set up continuous integration and continuous delivery (CI/CD) pipelines: Codifying end-to-end MLOps Directed Acyclic Graphs (DAGs) using Amazon SageMaker Pipelines (
ProcessingStep,TrainingStep,TransformStep,TuningStep,ConditionStep,LambdaStep,RegisterModel). Integrating with AWS Step Functions, Amazon EventBridge, and AWS CodePipeline for continuous integration and continuous deployment (CI/CD).
Domain 4: ML Solution Monitoring, Maintenance, and Security (24%)
- Task Statement 4.1: Monitor model performance: Setting up Amazon SageMaker Model Monitor across four drift dimensions: Data Quality drift (baseline statistics vs. live requests), Model Quality drift (merging S3 inference data with delayed ground truth labels to track metric degradation), Model Bias drift, and Feature Attribution (SHAP) drift. Configuring CloudWatch alarms and automated retraining triggers.
- Task Statement 4.2: Monitor and optimize infrastructure and costs: Tracking infrastructure performance metrics (utilization, throughput, availability, fault tolerance) with Amazon CloudWatch metrics, dashboards, and Logs Insights plus AWS X-Ray tracing; auditing API activity and invoking retraining with AWS CloudTrail; rightsizing instance families with SageMaker Inference Recommender and AWS Compute Optimizer; and governing spend with AWS Cost Explorer, AWS Budgets, cost-allocation tagging, and purchasing options (Spot, On-Demand, Savings Plans).
- Task Statement 4.3: Secure AWS resources: Enforcing IAM least-privilege execution roles,
iam:PassRoleconstraints, VPC endpoint configurations (AWS PrivateLink for SageMaker API, Runtime, and Notebooks), S3 and EBS encryption using AWS KMS Customer Managed Keys (CMKs), network isolation (EnableNetworkIsolation=True), and security auditing with AWS CloudTrail.
[!NOTE] Responsible-AI evaluation — pre-training and post-training bias with SageMaker Clarify, SHAP explainability, Model Cards, and Amazon Bedrock Guardrails — is assessed through Tasks 1.3, 2.3, and 4.1 (bias and explainability drift) and is taught in Chapters 11 and 12 of this guide. Cost governance is taught in Chapter 13.
3. Question Formats & Cognitive Complexity
The MLA-C01 exam features several distinct question structures designed to test both conceptual knowledge and operational troubleshooting ability.
| Question Format | Structure | Examination Mechanics |
|---|---|---|
| Multiple Choice (Single Response) | 1 correct option out of 4 choices | Tests specific service capabilities, parameter configurations, or optimal architecture selection under a specific constraint. |
| Multiple Response (Select Two or Three) | 2 correct out of 5 choices, or 3 correct out of 6 choices | Evaluates multi-component architectural designs. No partial credit is awarded; all correct options must be selected. |
| Scenario / Sequence Items | Paragraph-length operational problem statement | Presents a realistic enterprise challenge with explicit optimization constraints (e.g., "with the LEAST operational overhead", "with the LOWEST latency", or "at the LOWEST cost"). |
| Troubleshooting & Root Cause | Architectural failure or error code description | Requires identifying the missing IAM policy permission, broken VPC security group, misconfigured Feature Store join, or incorrect Model Monitor baseline file. |
4. Certification Ecosystem Comparison
To position your preparation accurately, review how MLA-C01 relates to other AWS credentials across the foundational, associate, and specialty tiers.
+-----------------------------------------------------------------------------+
| AWS AI & ML CERTIFICATION TAXONOMY |
| |
| [SPECIALTY] (MLS-C01 - RETIRED MARCH 2026) |
| - Heavy mathematical ML algorithms & legacy modeling |
| | |
| v |
| [ASSOCIATE TIER] AWS Certified Machine Learning Engineer (MLA-C01) |
| - Core Focus: Production MLOps, SageMaker Pipelines, |
| Feature Store, Model Registry, Serving, Monitoring |
| |
| [ADJACENT ASSOC] AWS Certified Data Engineer — Associate (DEA-C01) |
| - Core Focus: Data Lakes, Glue ETL, Redshift, EMR, |
| Streaming Ingestion (Kinesis/MSK), Lake Formation |
| |
| [FOUNDATIONAL] AWS Certified AI Practitioner (AIF-C01) |
| - Core Focus: High-level GenAI concepts, Bedrock, |
| prompt engineering, responsible AI fundamentals |
+-----------------------------------------------------------------------------+
Detailed Certification Comparison Table
| Certification | Level | Primary Target Audience | Core Focus Areas | Overlap with MLA-C01 |
|---|---|---|---|---|
| MLA-C01 (This Exam) | Associate | ML Engineers, MLOps Engineers, Applied Data Scientists | Production SageMaker, Pipelines, Feature Store, Model Monitor, Clarify, Inference Endpoints, Bedrock engineering | 100% (The target curriculum) |
| AIF-C01 | Foundational | Business stakeholders, Product Managers, Novice Cloud Engineers | High-level GenAI terminology, Bedrock foundational capabilities, prompt engineering concepts, responsible AI overview | ~20% (AIF-C01 is conceptual; MLA-C01 requires hands-on SDK, IAM, and infrastructure depth) |
| DEA-C01 | Associate | Data Engineers, Big Data Architects | S3 lakehouses, AWS Glue ETL, Amazon Redshift, EMR Spark, Athena, Kinesis, Lake Formation governance | ~35% (Strong overlap in Domain 1 Data Ingestion, S3 storage tiers, Glue ETL, and Cataloging) |
| SAA-C03 | Associate | Solutions Architects, System Administrators | VPC networking, EC2 auto-scaling, S3, IAM policies, RDS, DynamoDB, resilient cloud design patterns | ~25% (Foundation for VPC endpoints, PrivateLink, IAM execution roles, and KMS security) |
| MLS-C01 (Retired March 31, 2026) | Specialty | Senior ML Scientists | Classical statistical ML algorithms, mathematical hyperparameter formulas, manual Scikit-learn workflows | Replaced by MLA-C01 (MLS-C01 lacked modern SageMaker Pipelines, Feature Store, and GenAI/Bedrock focus) |
5. Strategic Exam-Taking & Elimination Frameworks
Success on the MLA-C01 exam requires disciplined time management and systematic elimination of distractors.
+-----------------------------------------------------------------------------+
| THE TWO-PASS EXAM PACING STRATEGY |
| |
| [PASS 1: 0 to 80 MINS] ---> Answer high-confidence questions immediately|
| (Target: 65 Questions) Flag lengthy scenario & complex multi-select|
| questions for second review |
| Pace: ~75 seconds per direct item |
| | |
| v |
| [PASS 2: 80 to 120 MINS] ---> Systematically work through ~15-20 flagged |
| (Target: Flagged Items) questions; dissect elimination keywords |
| Pace: ~2.5 to 3.0 minutes per flagged item |
| | |
| v |
| [FINAL REVIEW: 120-130 M]---> Ensure 0 unanswered questions |
| Confirm no accidental click changes |
+-----------------------------------------------------------------------------+
The Keyword Decoding Matrix (Constraint Filters)
AWS scenario questions contain specific qualifier phrases that instantly rule out multiple answer choices:
-
"With the LEAST operational overhead" / "Serverless solution":
- Choose: Managed SageMaker services (SageMaker Serverless Inference, SageMaker Data Wrangler, AWS Glue Serverless, Amazon Bedrock).
- Eliminate: Self-managed EC2 GPU clusters, self-hosted MLflow on EC2, custom Kubernetes (EKS) operators when native SageMaker features exist.
-
"Sub-second real-time inference with consistent high traffic":
- Choose: SageMaker Real-Time Endpoint with provisioned instances (e.g.,
ml.c6i.xlargeorml.g5.xlarge) and Target Tracking Auto Scaling. - Eliminate: Serverless Inference (unacceptable cold-start latency) or Asynchronous Endpoints.
- Choose: SageMaker Real-Time Endpoint with provisioned instances (e.g.,
-
"Large inference payload (e.g., 200 MB audio/video files) with processing time up to 15 minutes":
- Choose: SageMaker Asynchronous Inference Endpoint (supports payloads up to 1 GB and 1-hour timeout; integrates with S3 and SNS).
- Eliminate: Real-Time Endpoint (payload limit is 6 MB, maximum timeout is 60 seconds).
-
"Prevent training-serving skew and eliminate target leakage in historical training sets":
- Choose: SageMaker Feature Store with
point_in_time_queryusing event timestamps on the Offline Store. - Eliminate: Ad-hoc SQL joins on raw transactional tables without timestamp alignment.
- Choose: SageMaker Feature Store with
-
"Cost-effective model training that can tolerate instance interruptions":
- Choose: SageMaker Managed Spot Training with S3 checkpointing enabled (
checkpoint_s3_uri). - Eliminate: On-Demand instances or uncheckpointed Spot training (which risks total progress loss upon reclaim).
- Choose: SageMaker Managed Spot Training with S3 checkpointing enabled (
A machine learning engineer is preparing to take the AWS Certified Machine Learning Engineer — Associate (MLA-C01) exam. Which statement accurately describes the scoring mechanism and question distribution of the examination?
An enterprise wants to build an automated feature pipeline that serves real-time fraud detection predictions with sub-10ms latency while also generating point-in-time accurate historical training datasets without feature leakage. Which AWS service directly addresses this dual requirement with the least operational overhead?
An ML engineer needs to train a large deep learning computer vision model on a multi-node GPU cluster. The budget is strictly constrained, but the training pipeline can save model checkpoints to Amazon S3 every 20 minutes. Which training configuration provides up to 90% cost savings while preventing data loss from instance interruptions?
A financial services organization must host an image document extraction model that processes variable-sized PDF scans ranging from 20 MB to 150 MB. Processing each document takes between 2 to 5 minutes. The application requires an architecture that can scale instance capacity down to zero during inactive periods to minimize costs. Which SageMaker inference deployment option should the ML engineer select?